Why Do Notes Disappear When You Change Sequencer Resolution? A Same-Row Collision Preflight

A coarser view, MIDI quantization, and cell resampling are different operations

“Set the grid to 1/16” can describe several unrelated actions. Apple documents a Logic Pro region-quantize parameter that changes playback positions non-destructively while retaining the original timing. Ableton separately lets an editor disable grid snapping and provides a Quantize MIDI Tool for selected notes. Steinberg moves selected MIDI-event starts toward grid positions and also documents “double notes”—same-pitch notes at exactly the same position—that can remain as separate events until a dedicated command removes them. Novation documents six microsteps inside one Circuit Tracks step, so one visible coarse step can contain repeated triggers. None of those storage models implies that one cell must equal one stored onset.

FingerDrum’s current step pattern is different: each instrument row and step address contains one StepCell. Changing the resolution computes a new total number of cells for the same beat-and-bar container, resamples every active cell into that new matrix, then commits the result. The UI compares active-cell counts before and after the remap; if the count falls, it displays the warning that dense notes may merge. That warning identifies a loss after calculation, but it does not name the row, source cells, target cell, or winning attributes. The preflight supplies those missing details.

Freeze P0 before touching the resolution picker

Duplicate the FingerDrum project and name the untouched copy P0. Work on a second copy named C0. The project library exposes a Duplicate action; use it before the change, not after the warning. A screenshot is useful for visual comparison, but it is not a substitute for the editable source because it does not preserve velocity or note length.

1. Write the container: record beats per bar, number of bars, source resolution, and target resolution. Do not change tempo, loop bars, row order, or sounds during this test.
2. Calculate grid sizes: FingerDrum offers 1/4, 1/8, 1/8T, 1/16, and 1/16T. Total cells equal steps per beat × beats per bar × bars.
3. Inventory active cells: open Instruments and make every existing row visible. Count lit cells from left to right, beginning with 01. On the grid, count each amber note bar from its start through its right edge; that actual source-cell span is oldLength. Tap the lit cell to open Edit Note and record its displayed velocity percentage. Treat any highlighted note-value chip only as a cross-check: a dragged length may match no chip, while rounding or end-of-pattern limiting can make more than one chip match. Count active cells per row and overall.
4. Choose one direction: this example audits 1/16 → 1/8 in one 4/4 bar, so O=16 and N=8. Do not combine a straight-to-triplet change with a second edit.
5. Do not switch yet: finish the address ledger first. The predicted count is the acceptance criterion for C0.

Map every source address before listening

FingerDrum iterates each row independently. In zero-based coordinates, source cell s maps to:

Repeat the arithmetic through the full pattern; do not restart the mapping at a screen page unless the stored pattern itself restarts there. A triplet-to-straight conversion needs its actual full-pattern totals. For example, a one-bar 1/16T → 1/16 change uses 24 → 16, not a guessed rule that simply deletes “every third triplet.”

Only groups with the same row and target address collide

Give each active source cell a key (row, target), then group identical keys. In this controlled 16 → 8 example, the hat row contains adjacent source cells 01 and 02. Both map to hat target 01, so that bucket is many-to-one. A snare at source 02 also maps to target 01, but its row is different; it remains a separate event and can sound simultaneously with the surviving hat.

The four source cells produce three unique target keys, so the expected active-cell count is three. This is a structural calculation, not an audio judgment. Choke groups, sample tails, and effects may make two surviving rows sound like one gesture; conversely, a long sample may hide the missing retrigger. Compare the matrix first and audio second.

Predict the surviving velocity and note length—not just the count

When a FingerDrum target bucket is empty, the first mapped source cell fills it. A later source cell replaces the existing one only if its velocity is higher. Therefore the (Hat,01) bucket above keeps source 02 at velocity 0.78. If velocities are equal, the earlier source cell remains because later equal values do not replace it. This is a deterministic implementation rule, not a recommendation that “louder is musically better.”

The surviving source length is then scaled in proportion to the grid-size change:

Run P0 → C0 → R1 and prove that finer resolution is not an undo

With the ledger complete, change only C0 from the source resolution to the target. The on-screen warning is expected only when at least one active cell is lost. Do not treat the warning itself as a pass: inspect every target key and attribute.

1. Count gate: C0 overall count must equal the number of unique (row, target) keys. Each row count must match its own unique-target count.
2. Address gate: every active C0 cell must appear in the source-to-target map; no source may be assigned with a nearest-grid rule or a page-local restart.
3. Attribute gate: every collision bucket must keep the predicted velocity survivor and its proportionally scaled, clamped length.
4. Boundary gate: play through the loop return. Confirm the last target cell and next downbeat without changing tempo, bar count, sounds, or effects.
5. Return test: duplicate C0 as R1 and switch R1 back to the original finer resolution. Compare R1 with P0 cell by cell.

R1 should not equal P0 after a collision. Upsampling maps each surviving coarse cell into a finer address, but it has no record that one coarse cell previously came from two source cells. In the example, the single surviving Hat target 01 can move back into a fine-grid position; the discarded neighboring hat does not reappear. This is why the pre-change duplicate is evidence, not housekeeping.

What should I do when the preflight finds a collision?

Choose deliberately before switching. Keep the finer grid when both onsets are required; choose one source event and document the approximation when only one coarse cell is allowed; or use a target whose manual confirms an appropriate microstep, retrigger, or multi-event representation. Stop the conversion if the target cannot state the intended rhythm. Do not call a one-hit substitute lossless.

Is this the same as quantizing a finger-drummed MIDI take?

No. The earlier FingerDrum quantization guide compares timing correction, amount, anchors, groove, and protected ornaments in a performance. This audit begins with an already valid step pattern and tests the App’s resolution-resampling function. It does not decide whether the source performance should sound more precise or more human.

Will simultaneous kick and snare merge?

Not in this FingerDrum operation merely because they share time. Rows are resampled separately, so (Kick,05) and (Snare,05) are different keys. A collision requires two or more active source cells from the same row to receive the same target cell.

Does switching from straight to triplet resolution only change density?

No. It also changes the coordinate partition. With one 4/4 bar, 1/16T → 1/16 maps 24 cells to 16; some adjacent triplet cells share a target while others do not. Calculate all 24 addresses. Do not describe the result as “remove every third hit,” and do not infer a universal tie-breaking rule from another DAW.